Integrating Site Characterization with Aquifer and Soil Remediation

Jan 24, 2002 - Design reliability is used to guide site exploration, to compare different remedial designs, and to indicate when sufficient data have ...
0 downloads 0 Views 2MB Size
Chapter 22

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

Integrating Site Characterization with Aquifer and Soil Remediation Design J e j u n g L e e , H o w a r d W. Reeves*, a n d C h a r l e s H. D o w d i n g Department of C i v i l Engineering, Northwestern University, 2145 Sheridan Road, Evanston, I L 60208-3109

This review describes various approaches to propagate uncertainty inherent with the hydrogeological, chemical, and biological input parameters for groundwater flow and contaminant transport models employed to design remediation systems. Design reliability is used to guide site exploration, to compare different remedial designs, and to indicate when sufficient data have been collected to evaluate the design. Also described is a computationally efficient framework to integrate site characterization with aquifer and soil remediation design. A simple example is used to illustrate how design reliability is developed and employed to assess both exploration and design.

Site characterization and source identification are crucial to successful design of systems to remediate soil and groundwater contamination. There are, however, several factors that make it difficult to predict contaminant movement, and, therefore, make it difficult to design a remediation system. First, the geologic structure is usually heterogeneous, but only a limited number of data are available to describe the heterogeneity. Second, it is often difficult to completely

384

© 2002 American Chemical Society Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

385 identify the initial contaminant source. Third, contaminant movement is impacted by chemical and biological reactions that both are chemical and site specific. While a variety of remediation techniques to mitigate groundwater and soil contamination have been developed, assessment of their effectiveness before construction has lagged. One of the most widely used mitigation techniques is the pump-and-treat method, which has been used at about three-quarters of the Superfund sites where groundwater cleanup was required (1). However, the effectiveness of conventional the pump-and-treat method to reach the restoration goal has been questioned (1). Recently, in-situ treatment techniques such as monitored natural attenuation, passive reactive barriers, and bioremediation have gained popularity for groundwater contaminant remediation. As in-situ treatment techniques are increasingly employed, the recognition of the importance of site characterization is also increasing. However, knowledge of geologic structure and accuracy of model prediction are still uncertain. These uncertainties give rise to model performance uncertainty that affects the remedial design reliability. Therefore the impact of uncertain hydrogeology on the model performance must be assessed during design. This paper reviews methods to assess design reliability and presents an efficient framework to both evaluate various remedial designs and integrate site characterization into the design process. Within this framework, whether there are sufficient data to base the design, what additional data are required, and where these data should be collected to increase confidence in the design are quantitatively addressed.

Overview The fundamental assertion of our research is that remediation design must be considered as site characterization activities progress to yield cost-effective design and effective site characterization. This assertion arises from the observation that different designs require different hydrological, chemical, and biological information from the site. Model predictions, either analytical or numerical, are generally used to design remediation schemes. Due to the lack of information of site, uncertain input parameters are used in the models, therefore, the predicted performance of the design is uncertain. Freeze et al. (2) present a hydrologie decision framework illustrating how various uncertainties can be taken into account by considering the reliability of design and the value of future data. Applicability of this framework was demonstrated in a series of papers (3-5). The approach presented herein builds upon this work.

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

386 Figure 1 is the basic framework of this methodology. As shown, it is possible to estimate the probability of success for a design based on the current field data and to assess if additional data will improve the reliability of the design. Information obtained through the procedure also indicates which data should be collected and where on the site it should be obtained.

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

Date Sampling

Input Parameter Model

Performance Model

Redesign Additional Sampling

Figure 1. Computational framework to incorporate geologic/chemical input uncertainty with model sensitivity to calculate performance reliability

Review Many approaches have been developed for each step of the assessment framework shown in Figure 1. Conventional and newly introduced methods based on typical deterministic models are categorized in Figure 2. In this section, each method will be briefly reviewed and discussed.

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

387

input Parameter Model

Bayes J

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

f

\

Kriging

V

Sensitivity Analysis

J

ι

Performance Uncertainly

Evaluation of Performance Reliability

Monte Carlo Simulation

Reliability Index

Point Estimate Method

FORM/SORM

Γ First-Order Second-Moment Method I

ι Direct Method

Figure 2. Flowchart of alternative and employed methods for modeling at each step of the framework given in Figure 1. Thick line represents theflowemployed in our framework

Input: Parameters and their uncertainties For the input step, there we consider two basic approaches to interpolate between known geologic data. The most popular method, kriging will be discussed first, followed by the Bayesian approach. Kriging is the geostatistical method to extrapolate or interpolate point values at unsampled locations within the spatial domain limited data samples. It provides an estimation variance that yields a measure of the uncertainty of its interpolated values. Traditionally, kriging takes the form of linear interpolation based on Gaussian condition. It includes simple kriging, ordinary kriging, universal kriging, and cokriging. Among them, ordinary kriging is the most widely used methods (6). Ordinary kriging is a "best linear unbiased estimate" of the parameter. It is "linear" because its estimates are weighted linear combinations of the available data, "unbiased" because it tries to have the mean error equal to zero, and "best" because it aims at minimizing the variance of the error (7). The kriging estimator, Z* is described as,

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

388 Ζ*=ΣΛ Ζ a α

where Z is the measured data, λ is an estimation weight whose summation should be a unity to satisfy the non-bias condition. Ordinary kriging is applied when the mean is constant and not known. If the mean is not known but is expressed as a polynomial function of spatial coordinate, universal kriging may be used. If the mean of variable is perfectly known, simple kriging is applicable. In most practical situation, however, the mean is not known. It can be possibly known only when the number of data becomes so large as to allow an estimate of the mean. In many cases, data with more than one variable are obtained, and sometimes they are correlated. Cokriging was developed as a multivariate generalization of kriging (8). It is especially advantageous in cases where the primary measurements are limited and expensive, while auxiliary measurements are available at low cost (9). Hoeksema and Kitanidis (10) compared Gaussian conditional mean and variance with extended cokriging estimations based on point observations of transmissivity and hydraulic head. Harvey and Gorelick (11) applied conditional cokriged estimation to different types of measurement to show that sequential conditioning of data improves linear approximation and brings about a better estimate of hydraulic conductivity. If the measured data show non-Gaussian distribution or highly skewed histograms, linear kriging is not applicable. For these data, indicator kriging was developed as a non-linear kriging method. In this procedure, the original values are transformed into indicator values, such that they are zero if the datum value is less than the pre-defined cutoff level or unity if greater. Therefore, the estimated value with indicator kriging represents the probability of non-exceedence at a location (9). a

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

(D

α

α

The Bayesian approach is beginning to come into favor since the importance of site characterization and sampling strategies has increased. This approach is similar to the geostatistical approach (kriging) in its use to develop statistically realistic descriptions of sites to aid in initial data collection and site exploration when there is a scarcity of hard data (6,12). The Bayesian approach may provide the general framework to update parameter uncertainty when additional data are available, and to evaluate its effect on estimation or decision (13). Many researchers have adopted the Bayesian concept for environmental problems. Kitanidis (13) described the Bayesian conditional mean and variance which can be a nonlinear estimator for uncertain covariance parameters. McLaughlin et al. (14) presented a stochastic site characterization procedure that

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

389 combined field measurements with predictions obtained from mathematical models. In their framework, the model provides prior estimates, and they are updated whenever new measurements of hydraulic conductivity, head, or concentration become available. James and Gorelick (15) suggested a Bayesian data worth framework to improve the cost-effectiveness of data collection in groundwater remediation programs. McLaughlin and Townley (16) offered a comprehensive review of the inverse problem and demonstrated how inverse problems may be cast in a Bayesian framework. They described that in most applications of a posteriori estimation, the prior and measurement error probability densities were assumed to be Gaussian. With a multivariate Gaussian assumption, the Gaussian conditional approach can be used to estimate mean and its covariance in Bayesian framework. It is useful even when the Gaussian assumption does not hold, because for known parameters, it gives the minimum variance estimate which is linear function of the observation (13). The work discussed here adopts a Gaussian conditional approach to estimate the means and the variance of uncertain input parameters to the design model. Omre (77), Omre and Halvorsen (18) and Handcock and Stein (19) presented a Bayesian-Kriging approach to use the spatial approximation techniques of kriging while allowing for the uncertainty in specification of the parameters in the geostatistical approach (17). A prior distribution of the expected model is combined with available observations to make a posterior prediction. With an uninformed prior calculation, Bayesian-Kriging is very close to kriging with variogram data. This Bayesian alternative to kriging was compared with cokriging by Le et al. (20) and Sun (21). They showed that the Bayesian alternative uses all the historical information whereas cokriging only uses the current information for its prediction. They found that in terms of the mean-squared prediction error, prediction by the Bayesian alternative is superior to cokriging although the difference is small. They also found that in estimating the standard errors associated with prediction, the Bayesian alternative is also superior to cokriging. 9

Design Model : Performance and uncertainty Next, it is necessary to use the design model to determine an estimate of system performance and the uncertainty in this estimate. As described in this section, this step requires a large amount of computing time by conventionally used Monte Carlo simulation. As an alternative, the first-order second moment (FOSM) method adopted within our framework reduces the computational burden.

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

390 Monte Carlo simulation (MC) is the most widely used method for calculation of performance variance. It can be applied when many classes of parameters are simultaneously uncertain, and it is easy to understand (22). This method employs the classical statistics to calculate the variance with a set of model outputs from a set of input parameters that are randomly generated. The number of runs depends on the model and the assumed input parameter distribution. According to Harr (23), the required number of M C simulations, N, for, m, independent variables is estimated as, N=(h /4^) , where h is the standard deviation in a normal distribution corresponding to the confidence interval, and ε is the maximum allowable system error in estimating the confidence interval. For example, if a required confidence interval is 99% with 1% system error, h is 2.58, ε is 0.01, and (16,641) is estimated. Therefore computing time is the major disadvantage of this method. Cawlfield and W u (24) required over 400,000 computer runs to achieve a good level of accuracy for a one-dimensional transport code for a reactive contaminant.

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

2

m

m

The point estimate method (PEM) was developed as an alternative to Monte Carlo simulation and first-order second moment (FOSM) method. In this method, the model is evaluated at a discrete set of points in the uncertain parameter space, with the mean/variance of model predictions computed using a weighted average of these functional evaluations (25). Compared with M C , this method has the advantage of low computational requirement, and compared to the more computationally efficient F O S M , this method has the advantage to handle nonlinear models while F O S M requires linear models. Harr (23) described applicability of this method for structural problems, and Mishra (25) illustrated this method with an analytical model of health risk arising from waterborne radionuclide migration from a repository. The first-order second moment method (FOSM) is the method adopted within the framework to propagate input parameter uncertainty through numerical models (26, 27). F O S M provides two moments, mean and variance of predicted variables. This method is based on Taylor series expansion, of which second-order and higher terms are truncated. The expected value of concentration, E[u] and its covariance, COV[u] are (23, 27),

E[u(z)] = u(z)

COV[u (z),u (z)]=l i

j

(2)

Σ ^L^IC0V(^^) k=l 1=1 ÔZ

k

(3)

OZi

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

391 £ίί is a Jacobian matrix that relates the change in computed concentration, u to

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

dz the change in given input parameter, z. Equation (2) states that the best estimate of concentrations at every point in the discritized domain depends on the best estimate of input parameters. The estimate of the covariance of concentration for all the computation points in the domain is given by Equation (3), and it is determined by using the input parameter covariances and the sensitivity. Wagner and Gorelick (28) used F O S M analysis to propagate parameter uncertainty into contaminant concentration prediction uncertainty for a hypothetical aquifer. James and Oldenburg (29) compared F O S M analysis with M C analysis for three dimensional T C E concentration uncertainty. The comparison indicates that for the large-scale, low concentration problem, the two methods of analysis give similar uncertainty evaluations allowing justification of the use of F O S M analysis for conceptual models. Tiedeman and Gorelick (30) examined the reliability of remediation design and optimal locations for pumping wells used to contain a contaminant plume in an aquifer with uncertain parameters. They used the F O S M to translate input parameter uncertainty into estimated simulation uncertainty and showed that it gave results similar to the M C simulation. McKinney and Loucks (31) applied F O S M to a network design algorithm for improving the reliability of groundwater simulation model predictions. They presented a method to minimize the prediction variance by choice of new aquifer measurement locations with this method. Sensitivity coefficients from F O S M approaches are essential to predict uncertainty of model performance and design reliability. A perturbation method is generally used method to obtain the sensitivity matrix. However, this method requires a great deal of computing time and a determination of the proper perturbing value. Yeh (32) reviewed three methods: influence coefficient method (33) based on the perturbation concept, sensitivity equation method, and variational method (34, 35). Skaggs and Barry (36) compared the direct method, which involves analytically differentiating the equation, with an adjoint method. In recent years, Graettinger (37), Kunstmann (38), and Graettinger and Dowding (39) introduced the direct computation of sensitivity derivatives within the numerical code rather than numerical approximation of these derivatives. B y using an automatic differentiation tool such as ADIFOR (40), this approach can be performed both accurately and efficiently (41, 42). The direct derivative coding method is consistent with the analysis of Yeh (32) and Skaggs and Barry (36) and has the advantage that it requires only a single model run to obtain the whole sensitivity matrix over the perturbation method which requires N + l model runs where Ν is the number of input variables.

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

392 Output : Performance reliability

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

Finally the expected value and variation of the performance must be evaluated. Some, as described below have developed the first- and second-order reliability methods for this task. The reliability index approach followed herein provides similar information but appears to be much more efficient. The first- and second-order reliability methods (FORM and SORM) were initially developed for structural reliability applications to estimate the occurrence of low-probability events. These methods are based on two main concepts: 1) formulation of a performance function describing the behavior of interest in terms of the random variables, and 2) transformation of the problem into standard normal space, where an estimate of probability is obtained (43). Through this method, the system reliability index is calculated. Jang et al. (44) also showed that sensitivity of the system reliability index identified which parameter has a major influence on the estimate of probability. While F O R M and S O R M were adopted as more practical alternatives to Monte Carlo techniques. The implementation by Jang et al. (44) is quite computationally expensive due to numerical (finite difference) approximation of required derivatives relating changes in the dependent variable to changes in the input parameters within the algorithms. The search for the design point also can be computationally intensive, especially for large numbers of random variables. Details of F O R M and S O R M are presented in the publications by Cawlfield and Sitar (43), Jang et al. (44), Der Kiureghian et al. (45), and Sitar et al. (46). The reliability index, β, may be calculated as an alternative for obtaining the design point from F O R M and S O R M . To evaluate the model performance, two measures are considered: the estimated variance in the dependent variable and reliability index β for the remediation design. β is defined as the difference between the calculated performance, «,· and the allowable performance, u u at given locations (x y h zd at a site divided by the standard deviation of the model behavior, a . It may be written as, a

owed

u

ui

β



MMowed

a

Uj

U i

B y examining the β values for specific target locations at a site, the reliability or probability of success of a modeled remedial design will be evaluated. The probability of success is given by Φ(β) where Φ is the cumulative normal distribution function. If the β values are near zero, the model remedial design implies that the concentrations will be near the allowable maximum

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

393 concentration. If the β is large and negative, the remedial system does not meet the desired performance, and if β is large and positive, the model predicts that the remedial scheme will be successful. Therefore, the engineer may examine β values and judge i f additional sampling may reduce the a and can improve system reliability or if the remedial system must be redesigned to change the expected performance, u . If the examination of the β shows reliability is not sufficient, geological uncertainty and model sensitivity are combined to determine where next data should be collected. The next location will be the data position that creates the greatest variance in the performance model. The estimated covariance matrix has been called the importancy matrix by several authors (37,47). ui

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

{

Discussion and Conclusion Various approaches to propagate uncertainty inherent with the hydrogeological, chemical, and biological input parameters for groundwater flow and contaminant transport models employed to design remediation system were reviewed and discussed. In terms of the framework presented in Figure 1, the results from reliability analysis may give information about (1) which data are required and where these data should be collected to increase confidence in the design, (2) which design alternative is most reliable, and (3) when the acceptable probability of design success is met. Lee (48) presented the application of the framework to two- and threedimensional finite element models for various design alternatives to answer the questions (1) through (3). Among methods that are presented in Figure 2, Bayesian conditional calculation for spatial geologic uncertainty, a first-order second moment (FOSM) calculation of model uncertainty, and reliability index calculation are adopted sequentially. For F O S M , the sensitivity matrix is calculated by directly differentiating numerical finite element code. The comparison of the performance uncertainties from multiple uncertain input parameters allowed a determination of the parameter that most influenced the performance model and was the most important for determination of the location for the next sample. Lee (48) demonstrated the use of the reliability index to compare the reliability of different proposed remediation design alternatives and to analyze the impact of additional sampling on the design and performance for the given cleanup goal at the compliance point. If more than one compliance point are examined, a multivariate β should be determined to consider correlation between the compliance points. More research is required in this area. Fortney (49) applied the framework to a study with field data. He adopted M O D F L O W 2000 (50, 51) as the groundwater flow model and incorporated the

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

394 calculation of performance uncertainty by F O S M . The influence of input parameter uncertainty on the performance uncertainty and the reliability of design alternatives for design goals at different locations were shown. Fortney (49) also presented an alternative way to determine the most influential point in the domain requiring the additional sampling to increase the reliability of the design. The work by Lee (48), Fortney (49), and Graettinger (37) present the details of integrating site characterization and design. A n important and interesting question that arises in the application of this framework is, What probability of success is sufficient ? Baecher (52) discussed socially acceptable probability of success for civil engineering projects, but no groundwater projects. To use our approach in practice, socially acceptable probability of success for groundwater remediation will need to be debated and determined.

Acknowledgement This work was funded by Grant R-827126-01-0 from the U.S. Environmental Protection Agency. The work has not been subjected to any E P A review, and therefore does not necessarily reflect the views of the Agency.

References 1.

2. 3. 4. 5. 6. 7.

Pump-and-Treat Ground-Water Remediation: A Guide for Decision Makers and Practitioners. U.S. Environmental Protection Agency (EPA), Office of Research and Development; Washington, D C 1996; EPA/625/R-95/005. Freeze, R . A . ; Massmann, J.; Smith, L.; Sterling, T.; and James, B . , Ground Water 1990, 28(5), 738-766. Massman, J.; Freeze, R.A.; Smith, L.; Sperling, T.; and James, B . , Ground Water 1991, 29(4), 536-548. Sperling, T.; Freeze, R.A.; Massman, J.; Smith, L.; and James, B . , Ground Water 1992, 30(3), 376-389. Freeze, R.A.; James, B . ; Massmann, J.; Sperling, T.; and Smith, L., Ground Water 1992, 30(4), 574-588. Chiles, J.-P.; Delfmer, P. Geostatistics: Modeling Spatial Uncertainty, John Wiley & Sons: New York, 1999. Isaaks, E . H . ; Srivastava, R . M . An Introduction to Applied Geostatistics; Oxford University Press: New York, 1989, 278-322.

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

395 8. 9.

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

23. 24. 25.

26. 27. 28. 29. 30. 31. 32. 33. 34.

Matheron, G . The Theory of Regionalized Variables and Its Applications; Ecole des Mines:Fontainebleau, France, 1971. Rouhani, S. Geostatistical Estimation: Kriging, In Geostatistics for Environmental and Geotechnical Applications; Srivastava, R.M.;Rouhani, S.; Cromer, M . V . ; Johnson, I.; Desbarats, A . J . Eds.; ASTM STP 1283; American Society of Testing and Materials, 1996, pp. 20-31. Hoeksema, R.J.; Kitanidis, P.K., Water Resour. Res. 1985, 21(6), 825-836. Harvey, C.F.; Gorelick, S.M., Water Resour. Res. 1995, 31(7), 1615-1626. Journel, A . G . ; Huijbregts, C.J. Mining Geostatistics; Academic Press: New York, 1978. Kitanidis, P.K., Water Resour. Res. 1986, 22(4), 499-507. McLaughlin, D . ; Reid, L . B . ; Li, S.-G.; Hyman, J., Ground Water 1993, 31(2), 237-249. James, B.R.; Gorelick, S.M., Water Resour. Res. 1994, 30(12), 3499-3513. M e Laughlin, D . ; Townley, L.R., Water Resour. Res. 1996, 32(5), 11311161. Omre, H . , Mathematical Geology 1987, 19(1), 25-39. Omre, H . ; Halvorsen, K.B., Math. Geology 1989, 21(7), 767-786. Handcock, M.S.; Stein, M . L . , Technometrics 1993, 35(4), 403-410. Le, N.D.; Sun, W.; Zidek, J.V., J.R.Statist.Soc.B 1997, 59(2), 501-510. Sun, W., Environmetrics 1998, 9, 445-457. Consequences of spatial variability in aquifer properties and data limitations for groundwater modelling practice; Peck, Α.; Gorelick, S.; de Marsily, G . ; Foster, S.; Kovalevsky, V . , Eds.; IAHS Publication No. 175; IAHS Press: Wallingford, Oxfordshire, U K , 1988. Harr, M . E . Reliability-Based Design in Civil Engineering; McGraw-Hill: New York, 1987. Cawlfield, J.D.; Wu, M., Water Resour. Res. 1993, 29(3), 661-671. Mishra, S. In Calibration and Reliability in Groundwater Modelling: Coping with Uncertainty; Stauffer, F.; Kinzelbach, W . ; Kovar, K . ; Hoehn, E . Eds.; IAHS Publication No. 265; IAHS Press: Wallingford, Oxfordshire, U K , 2000. Dettinger, M . D . ; Wilson, J.L., Water Resour. Res. 1981, 17(1), 149-161. Townley, L.R.; Wilson, J.L., Water Resour. Res. 1985, 21(12), 1851-1860. Wagner, B.J.; Gorelick, S., Water Resour. Res. 1987, 23(7), 1162-1174. James, A . L . ; Oldenburg, C . M . , Water Resour. Res. 1997, 2495-2508. Tiedeman, C.; Gorelick, S.M., Water Resour. Res. 1993, 29(7), 2139-2153. McKinney, D.C.; Loucks, D.P., Water Resour. Res. 1992, 28(1), 133-147. Yeh, W.W.-G., Water Resour. Res. 1986, 22(2), 95-108. Becker, L . ; Yeh, W.W.-G., Water Resour. Res. 1972, 8(4), 956-965. Carter, R.D.; Kemp, Jr, L.F.; Pierce, A . C . ; Williams, D.L., Soc.Pet.Eng.J. 1974, 14(2), 186-196.

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.

Downloaded by UNIV OF CALIFORNIA SAN DIEGO on July 15, 2016 | http://pubs.acs.org Publication Date: January 24, 2002 | doi: 10.1021/bk-2002-0806.ch022

396 35. Sun, N.Z.; Yeh, W.W.-G., Water Resour. Res. 1985, 21(6), 869-883. 36. Skaggs, T.H.; Barry, D.A., Water Resour. Res. 1996, 32(8), 2409-2420. 37. Graettinger, A.J. Ph.D. Dissertation, Northwestern University, Evanston, IL., 1998. 38. Kunstmann, H.G. Ph.D. Dissertation, Swiss Federal Institute of Technology, Zurich, Switzerland, 1998. 39. Graettinger, A . J . ; Dowding, C . H . , Journal of Geotechnical and Geoenvironmental Engineering, A S C E 1999, 125(11), 959-967. 40. Bischof et al., IEEE Comp. Sci. & Engng. Fall, 1996, 18-32. 41. Abate, J.; Wang, P.; Sepehrnoori, K.; Dawson, C., Commun. Numer. Meth. Engng. 1999, 15, 423-434. 42. Park, S.K.; Droegemeier, K . K , Mon. Wea. Rev. 1998, 127, 2180-2196. 43. Cawfield J.D.; Sitar, N. In Consequences of spatial variability in aquifer properties and data limitations for groundwater modelling practice; Peck, Α.; Gorelick, S.; de Marsily, G . ; Foster, S.; Kovalevsky, V . Eds.; IAHS Publication No. 175; IAHS Press: Wallingford, Oxfordshire, U K , 1988; pp. 191-216. 44. Jang, Y.-S.; Sitar, N.; Der Kiureghian, Α., Water Resour. Res. 1994, 30(8), 2435-2448.Der Kiureghian, Α.; L i n , H.-Z.; Hwang, S.-J., Journal of Engineering Mechanics, A S C E 1987, 113(8), 1208-1225. 45. Der Kiureghian, Α.; L i u , P.-L., Journal of Engineering Mechnics, 1986, 112(1), 85-104. 46. Sitar, N.; Cawlfield, J.D.; Der Kiureghian, A.D., Water Resour. Res. 1987, 23(5), 794-804. 47. Tomasko, D . M . ; Reeves, M . ; Kelley, V . A . In Proceedings of the DOE/AECL Conference on Geostatistical Sensitivity, and Uncertainty Methods for Ground-Water Flow and Radionuclide Transport Modeling; Buxton, B., Ed.; Battelle Press: Columbus, Ohio, 1987. 48. Lee, J. Ph.D. Dissertation, Northwestern University, Evanston, IL., 2001. 49. Fortney, M . D . M . S . Dissertation, Northwestern University, Evanston, IL., 2001. 50. Harbaugh, A.W.; Banta, E.R.; H i l l , M . C . ; McDonald, M . G . , MODFLOW­ -2000, the U.S. Geological Survey modular ground-water model - User guide to modularization concepts and the Ground-Water Flow Process: U.S. Geological Survey Open-File Report 00-92, 2000. 51. H i l l , M . C . ; Banta, E.R.; Harbaugh, A.W.; Anderman, E.R., MODFLOW­ -2000, the U.S. Geological Survey modular ground-water model - User guide to the Observation, Sensitivity, and Parameter-Estimation Processes and three post-processing programs, U.S. Geological Survey Open-File Report 00-184, 2000. 52. Baecher, G . B . , Geotechnical Risk Analysis User Guide, FHWA/RD-87011, Fed.Hwy.Admin.,McLean, 1987, p.55.

Lipnick et al.; Chemicals in the Environment ACS Symposium Series; American Chemical Society: Washington, DC, 2002.